MCQs - Model Validation & Evaluation

Why is model validation necessary after training a machine learning model?

• To reduce the size of the dataset

• To estimate how the model performs on unseen data

• To increase the number of training samples

• To remove noise from the dataset

In K-Fold Cross Validation, what happens in each iteration?

• The entire dataset is used only for testing

• One fold is used for testing while the remaining folds are used for training

• All folds are used simultaneously for training and testing

• Only the first fold is used for testing

Which cross-validation technique is most suitable for imbalanced datasets?

• Leave-One-Out Cross Validation

• Shuffle Split

• Stratified K-Fold Cross Validation

• Random Train/Test Split

In Leave-One-Out Cross Validation (LOOCV), how many folds are used if the dataset contains N samples?

• N/2

• N

• N – 1

• 2N

Why is Shuffle-Split Cross Validation not recommended for time-series data?

• It requires too much memory

• It changes the dataset size

• It breaks the temporal order of the data

• It increases model bias

Which of the following describes the accuracy paradox?

• Accuracy becomes zero when predictions are random

• A model can achieve high accuracy even if it has no real predictive power

• Accuracy is always lower than precision

• Accuracy cannot be used in classification problems

Which metric answers the question: “Out of all predicted positive cases, how many were actually positive?”

• Accuracy

• Recall

• Precision

• Specificity

Which metric answers the question: “Out of all actual positive cases, how many were correctly identified?”

• Recall

• Precision

• Specificity

• Accuracy

Which evaluation metric considers both precision and recall equally?

• Accuracy

• Specificity

• F1-Score

• Negative Predictive Value

If a classifier always predicts the negative class, which metric may still appear high in imbalanced datasets?

• Recall

• Precision

• Accuracy

• F1-Score